| Soil is the material basis for human survival and development,and is also the basic source of life for human beings.Coal mining,though contributing a lot to our society,has yielded an increasingly large area of mine goafs,resulting in land falling and collapsing,loss of water and soil erosion.In recent years,with the implementation of land reclamation project,the land in coal-mining areas has gradually recovered after comprehensive treatment.Hyperspectral remote sensing technology for the study of quantitative remote sensing is often used to determine soil nutrients content quickly and accurately,thus providing an important means for the monitoring and assessment of soil quality.In this study,the reclamation cropland in the coal mining areas of the loess plateau was chosen as the experimental area in Xiangyuan County,Shanxi Province.Soil samples were collected from the cropland and these samples were highly representative for the measurement of main soil nutrients.Under the laboratory conditions,the soil nutrients were measured based on hyperspectral analysis.And then such soil nutrients as organic matter,total nitrogen,available phosphorus and potassium in soil were taken as research objects.Their spectral properties were analyzed.Meanwhile,the spectral reflectance of the soil samples was measured and 3 different kinds of transformation were carefully analyzed.At last,unary linear regression method,multiple linear regression method and partial least squares regression(PLSR)method were used to build hyperspectral inversion models for this study area.(1)The results showed that the wave bands of spectral properties of main soil nutrients were determined.The significant bands of soil organic matter content(SOMC)were that the bands of the spectral reflectance(R)were 400-1,800 nm and 1,880-2,400 nm;the bands of first order differential reflectance(D(R)),420-790 nm,1,020-1,040 nm and 2,150-2,200 nm;the bands of inverse-log reflectance(1g(1/R)),400-1,830 nm and 1,860-2,400 nm.The maximum bands of absolute value concerning the correlation coefficients between the spectral indices and SOMC were that the band of the spectral reflectance(R)was 800 nm;the band of first order differential reflectance(D(R)),600 nm;the band of inverse-log reflectance(1g(1/R)),760 nm.The significant bands of total nitrogen were that the bands of the spectral reflectance(R)were 400-1,835 nm and 1,872-2,400 nm;the bands of first order differential reflectance(D(R)),420-760 nm,771-790 nm and 2,172-2,197 nm;the bands of inverse-log reflectance(1g(1/R)),400-1,837 nm and 1,874-2,400nm.The maximum bands of absolute value concerning the correlation coefficients between the spectral indices and total nitrogen content were that the band of the spectral reflectance(R)was 762 nm;the band of first order differential reflectance(D(R)),557 nm;the band of inverse-log reflectance(1g(1/R)),765 nm.The significant bands of available phosphorus were that the bands of the spectral reflectance(R)were 400-1,830 nm and 1,860-2,400 nm;the bands of first order differential reflectance(D(R)),403-763 nm,770-804 nm,1,197-1,301 nm and 2,172-2,197 nm;the bands of inverse-log reflectance(1g(1/R)),400-1,840 nm and 1,860-2,400nm.The maximum bands of absolute value concerning the correlation coefficients between the spectral indices and available phosphorus content were that the band of the spectral reflectance(R)was 1,340 nm;the band of first order differential reflectance(D(R)),565 nm;the band of inverse-log reflectance(1g(1/R)),1,335 nm.The significant bands of available potassium were that the bands of the spectral reflectance(R)were 400-1,834 nm and 1,817-2,368 nm;the bands of first order differential reflectance(D(R)),486-763 nm;the bands of inverse-log reflectance(1g(1/R)),400-1,835 nm and 1,874-2,400nm.The maximum bands of absolute value concerning the correlation coefficients between the spectral indices and available potassium content were that the band of the spectral reflectance(R)was 746 nm;the band of first order differential reflectance(D(R)),597 nm;the band of inverse-log reflectance(1g(1/R)),770 nm.(2)The prediction models of main soil nutrients content were established by using unary linear regression method.The values of R2 and RMSE of the R model of SOMC were 0.51 and 5.32 respectively;the values of R2 and RMSE of the D(R)model,0.67 and 4.41 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.53 and 5.25 respectively.The values of R2 and RMSE of the R model of total nitrogen content were 0.64 and 0.22 respectively;the values of R2 and RMSE of the D(R)model,0.67 and 0.21 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.64 and 0.22 respectively.The values of R2 and RMSE of the R model of available phosphorus content were 0.11 and 2.38 respectively;the values of R2 and RMSE of the D(R)model,0.20 and 2.36 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.11 and 2.37 respectively.The values of R2 and RMSE of the R model of available potassium content were 0.12 and 26.43 respectively;the values of R2 and RMSE of the D(R)model,0.13 and 26.30 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.19 and 25.38 respectively.(3)The prediction models of main soil nutrients content were established by using multiple linear regression method.The values of R2 and RMSE of the R model of SOMC were 0.80 and 3.46 respectively;the values of R2 and RMSE of the D(R)model,0.71 and 4.12 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.84 and 3.06 respectively.The values of R2 and RMSE of the R model of total nitrogen content were 0.73 and 0.19 respectively;the values of R2 and RMSE of the D(R)model,0.83 and 0.15 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.80 and 0.16 respectively.The values of R2 and RMSE of the R model of available phosphorus content were 0.41 and 1.96 respectively;the values of R2 and RMSE of the D(R)model,0.51 and 1.79 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.35 and 2.05 respectively.The values of R2 and RMSE of the R model of available potassium content were 0.44 and 21.39 respectively;the values of R2 and RMSE of the D(R)model,0.44 and 21.39 respectively;the values of R2 and RMSE of the 1g(1/R)model,0.35 and 22.90 respectively.(4)The prediction models of SOMC and total nitrogen content were established by using partial least squares regression(PLSR)method.Based on the full bands,the values of R2,RMSE and RPD of the R model of SOMC were 0.79,3.64 and 2.10 respectively;the values of R2,RMSE and RPD of the D(R)model,0.61,5.43 and 1.41 respectively;the values of R2,RMSE and RPD of the 1g(1/R)model,0.79,3.53 and 2.17 respectively.Based on the significant bands,the values of R2,RMSE and RPD of the R model of SOMC were 0.85,3.00 and 2.52 respectively;the values of R2,RMSE and RPD of the D(R)model,0.83,3.20 and 2.39 respectively;the values of R2,RMSE and RPD of the 1g(1/R)model,0.85,3.00 and 2.56 respectively.Based on the full bands,the values of R2,RMSE and RPD of the R model of total nitrogen content were 0.61,0.23 and 1.51 respectively;the values of R2,RMSE and RPD of the D(R)model,0.78,0.18 and 1.90 respectively;the values of R2,RMSE and RPD of the 1g(1/R)model,0.68,0.21 and 1.68 respectively.Based on the significant bands,the values of R2,RMSE and RPD of the R model of total nitrogen content were 0.70,0.20 and 1.77 respectively;the values of R2,RMSE and RPD of the D(R)model,0.79,0.17 and 2.04 respectively;the values of R2,RMSE and RPD of the 1g(1/R)model,0.69,0.20 and 1.74 respectively.(5)The optimal prediction models of main soil nutrients content were established for the study area.The optimal prediction model of SOMC was the 1g(1/R)model established by using partial least squares regression(PLSR)method.The values of R2,RMSE and RPD of the model were 0.85,3.00 and 2.56 respectively.The optimal prediction model of total nitrogen content was the D(R)model established by using multiple linear regression method.The values of R2 and RMSE of the model were 0.83 and 0.15 respectively.It could be concluded that the prediction accuracies of the hyperspectral inversion models for SOMC and total nitrogen content were much better while the prediction models for available phosphorus and potassium were not satisfactory. |
PDF Full Text Request